Skip to content

Latest commit

 

History

History
32 lines (22 loc) · 2.51 KB

README.md

File metadata and controls

32 lines (22 loc) · 2.51 KB
Raw

OutliersGAN

This code implements a one-dimensional Generative Adversarial Network which performs the following tasks:

  • It generates realistic 1D data distributions after training. For the case of my project, the data consisted of SDSS spectra;

  • It uses the trained model to generate an arbitrarily large number of fake spectra;

  • It access the layers of the discriminator after training in order to show that it is possible to perform outlier detection with GANs.

Notes

The following models were trained and are currently saved in OzStar:

  • legacy data; extra loss term; grid; valid: 391672
    name: wgangp_grid_newloss code: 71

  • galaxy with STARFORMING and STARBURST plus zWarning=0; extra loss term; grid; valid: 321520 # name: gal_zWarning code: 72

  • qso plus zWarning=0; extra loss term; grid; valid: 22791
    name: qso_zWarning co code 73

  • qso plus zWarning=0; extra loss term; grid; valid: 2346 different wavelength range: 1800-4150 Angstroms
    name: qso2_zWarning
    code: 74

Acknowledgements

I thank my supervisor, Dr. Karl Glazebrook, for having given both the opportunity and funding to work in such an interesting project during 5 months. I further thank Dr. Colin Jacobs for being so nice and helpful during my stay. I finally thank Swinburne University of Technology, and particularly the Centre for Astrophysics and Supercomputing (Melbourne, Australia) for welcoming me extremely well.